Multiple instance learning (MIL) has become a popular approach in computational pathology and whole slide image (WSI) analysis for their weakly-supervised nature. The introduction of attention-based instance pooling in particular has enabled enhanced interpretation of both model decision-making and the underlying data by leveraging attention weights. However, the interpretation of the attention weights, specifically its indication of instance classes, have been contested. We demonstrate that noise or heterogeneity in bagged data can require MIL classifiers to predict bag class based on prevalence of the positive instance class, as opposed to its presence; altering the behavior of the attention mechanism and likely contributing to the aforementioned discrepancies of attention weight interpretation. Here, we introduce an approach to identify and score instances which contribute to a positive bag label, robust against altered attention behavior, in two discreet settings. First, we elucidate how the behavior of attention-based pooling is altered, using the MNIST dataset, where bagged datasets are generated with a different threshold (t) of positive class instances defining the bag label. While maintaining a high bag-level classification score, the distribution of attention weights between positive and negative changed with t, where the mechanism attended more to positive instances for lower values of t, but favored negative instances for high values of t. We also apply our method to an in-house dataset of prostate cancer nuclei to predict the aggressiveness of the disease, and demonstrate how our method may be used to identify a subgroup of nuclei more highly associated with aggressiveness.

Breast cancer is the leading cause of cancer-related deaths among women worldwide. It is standard practice for patients to undergo a sentinel lymph node biopsy (SLNB) with breast surgery for staging. However, more than 60% of patients with primary operable breast cancers do not have axillary lymph node metastases. Hence, most patients risk complications from SLNB, but do not benefit. There has been considerable research into non-invasive methods for axillary lymph node staging, but few are externally validated. We hypothesized that a new method using large-scale DNA organization (LDO) analysis on a biopsy of the breast primary could predict axillary metastases. LDO, which measures various nuclear characteristics such as the size, shape, and chromatin texture, has been correlated with a variety of clinical outcomes in different cancers, including survival in breast cancer. In this study, we determined that Random Forest models performed best for LDO analysis. On an external test set, our models using LDO features alone achieved an area under the curve (AUC) = 0.711 and clinicopathological features alone achieved AUC = 0.741. Our best and final model using LDO and clinicopathological features together achieved AUC = 0.775. We identified several LDO features that were important for predicting axillary lymph node metastases, including nuclear radius and staining intensity. We present the first histologically-based, externally-validated, and explainable machine learning model capable of predicting axillary lymph node metastases preoperatively.

Head and neck squamous cell carcinoma (HNSCC) evolves through stepwise clonal expansion within genetically altered mucosa fields, yet actionable biomarkers remain undefined. Leveraging Fanconi anemia (FA), a cancer predisposition syndrome with extreme HNSCC risk due to defective DNA interstrand crosslink repair, we profiled premalignant changes in the oral cavity using noninvasive brush biopsies. Consistent with our prior demonstration of genomic instability in FA-associated SCCs, we detected pathogenic TP53 variants in 26% and copy number alterations in 60.5% in clinically normal-appearing oral mucosa of individuals with FA. These subclinical clonal expansions define candidate biomarkers of early clonal evolution amenable to serial sampling for risk stratification and prevention studies. Since FA-associated SCCs share genomic features with sporadic HNSCC, these findings may extend to the broader population. We also identify somatic reversion of a pathogenic FANCB variant, providing evidence of genomic self-correction and suggesting a potential avenue for gene-based cancer prevention in FA. Statement of significance:Oral mucosa of individuals with Fanconi anemia contains frequent abnormal clones creating a premalignant field that increases cancer risk. The noninvasive brush sampling approach allows repeated measurements, ongoing surveillance, and assessment of prophylactic strategies that may be useful in the prevention of cancers in people with FA and in the general population. Somatic reversion of a pathogenic FANC variant may protect the oral mucosa from DNA repair deficiency and premalignant clonal evolution.

Multiple instance learning (MIL) has become a popular approach to analyze histopathology datasets due to its weakly supervised nature. In particular, attention-based models can learn key instances of which labels are usually unknown or unavailable. For instance, large-scale DNA organization (LDO) analysis aims to make patient prognoses from quantitative features of nuclear morphometry and chromatin condensation calculated from images of the nucleus. Leveraging attention mechanisms allows a model to identify nuclei with alterations which likely contribute to patient outcome without individual cell labels. However, a crucial assumption of MIL that is often overlooked in histopathology applications, wherein a bag is positive if it has at least one positive instance. In a cancer context, this assumption is not robust, as a single malignantly transformed cell may be necessary but insufficient to cause carcinogenesis or malignant cancer progression. A reasonable adjustment is to learn a tolerable threshold of aberrant cells. The attention mechanism can be modified so that both key aggressive and indolent nuclei are identified in contrast to traditional MIL attention mechanisms which only give weight to positive instances. We demonstrate that this is an effective approach for prostate cancer (PCa) prognosis. A binary MIL classifier was trained to identify PCa patients (bags) of the indolent (negative) and aggressive (positive) outcomes. The cohort includes 38 Gleason score (GS) 6 patients who did not display signs of progression during active surveillance (AS) and 22 patients with GS 9 who died within 2 years of consultation. A linear attention layer identifies aggressive and indolent nuclei (instances) to generate a weighted mean representation of LDO features for the patient, reducing the influence of nuclei of ambiguous labels. To mitigate overfitting, weights are shared between the attention layer and patient classification layer and trained to optimize a combination of binary cross entropy loss on the nuclear and patient level. Patients in the training set were classified with a balanced accuracy of 0.851, and an F1-score of 0.815. This performance also translated to the patients in the holdout set, where the balanced accuracy and F1-scores of patient classification was 0.857 and 0.833 respectively. Tests on an independent cohort of 147 patients with GS 7+ also demonstrate that LDO score is correlated with GS, biochemical recurrence following brachytherapy, and progression in GS6 active surveillance patients. Future studies will analyze the performance of the trained classifiers on patients with GS7 and GS8 further, such as the classifier’s ability to rank severity of clinical outcomes with c-index. These results demonstrate the potential of the MIL-based LDO biomarker for prostate cancer patient prognosis and management. Further validation on the brachytherapy-treated cohort, and survival analysis will be done to assess the performance of the MIL classifier, and its potential benefit for prostate cancer management. Fumiya Inaba, Zhaoyang Chen, Anita Carraro, Paul Gallagher, Mira Keyes, Martial Guillaud, Calum MacAulay. Multiple instance learning of large-scale DNA organization to characterize prostate cancer aggressiveness [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr B030.

Supplemental Table 2: Integration assay primer and probe sequences

Artificial intelligence (AI) is becoming an integral part of pathological assessment and diagnostic procedures in modern pathology. As most prostate cancers (PCa) arise from glandular epithelial tissue, an AI-based methodology has been developed to recognize glandular epithelial nuclei in prostate biopsy tissue. An integrated machine-learning network, named GlandNet, was developed to correctly recognize the epithelial cells within prostate glands using cell-centric patches selected from the core biopsy specimens. Feulgen-Thionin (a DNA stoichiometric label) was used to stain biopsy sections (4–7 µm in thickness) from 82 active surveillance patients diagnosed with PCa. Images of these sections were human-annotated, and the resultant dataset consisted of 1,264,772 segmented, cell-centric nuclei patches, of which 449,879 were centered on epithelial gland nuclei from 110 needle biopsies (training set: n = 66; validation set: n = 22; and test set: n = 22). The training of GlandNet used semi-supervised machine-learning knowledge of the training and validation cohorts and integrated both human and AI predictions to enhance its performance on the test cohort. The performance was evaluated against a consensus deliberation from three observers. The GlandNet demonstrated an average accuracy, sensitivity, specificity, and F1-score of 94.1%, 95.7%, 87.8%, and 95.2%, respectively, when tested on the 20,735 glandular cells found in the three needle biopsies with the visually best consensus predictions. Conversely, the average accuracy, sensitivity, specificity, and F1-score were 90.9%, 86.4%, 94.0%, and 89.7% when assessed on 57,217 cells found in the three needle biopsies with the visually worst consensus predictions. GlandNet is a first-generation AI with an excellent ability to differentiate between epithelial and stromal nuclei in core biopsies from patients with early prostate cancer.

Diagnosis and classification of oral epithelial dysplasia (OED) is critical to identifying and prognosticating patients at risk of squamous cell carcinoma (SCC). However, conventional 3-tiered and 2-tiered grading systems suffer from poor inter-pathologist agreement, and SCC may arise from all grades of OED. This study evaluated pathologist agreement in OED classification as p53 wildtype, p53 abnormal, and HPV-associated based on recent evidence demonstrating the utility of p53/p16 immunohistochemistry (IHC) in this setting and increased risk of p53 abnormal OED progression to SCC, regardless of histologic grade. Fifty digital biopsy specimens were evaluated for diagnosis by 18 subspecialty-trained pathologists, with OED graded utilizing 3-tiered, 2-tiered, and p53 wildtype/p53 abnormal/HPV-associated schemata. Cases were reviewed first without and subsequently with p53/p16 IHC. The cohort consisted of 8 cases of p53 wildtype, 24 cases of p53 abnormal, and 18 cases of HPV-associated OED. Inter-pathologist agreement in OED grading according to 3-tiered (κ=0.32) and 2-tiered (κ=0.39) systems by H&E was poor, but fair-to-good (κ=0.59) in classification as p53 wildtype/p53 abnormal/HPV-associated by H&E and IHC. Classification of OED as p53 wildtype, p53 abnormal, or HPV-associated using p53/p16 IHC outperformed conventional grading in this cohort enriched for p53 abnormal OED, which required correct interpretation of p53 IHC, historically deemed challenging. Routine use of IHC also identifies a wider histologic spectrum of HPV-associated OED than is currently appreciated. More work is needed to determine the efficacy of this classification system in predicting patient outcomes and in guiding management decisions in real-world cohorts.

Supplemental Table 1: Tag SNPs and Calculated Odds Ratios

Diagnosis and classification of oral epithelial dysplasia (OED) is critical to identifying and prognosticating patients at risk of squamous cell carcinoma (SCC). However, conventional 3-tiered and 2-tiered grading systems suffer from poor inter-pathologist agreement, and SCC may arise from all grades of OED. This study evaluated pathologist agreement in OED classification as p53 wildtype, p53 abnormal, and HPV-associated based on recent evidence demonstrating the utility of p53/p16 immunohistochemistry (IHC) in this setting and increased risk of p53 abnormal OED progression to SCC, regardless of histologic grade. Fifty digital biopsy specimens were evaluated for diagnosis by 18 subspecialty-trained pathologists, with OED graded utilizing 3-tiered, 2-tiered, and p53 wildtype/p53 abnormal/HPV-associated schemata. Cases were reviewed first without and subsequently with p53/p16 IHC. The cohort consisted of 8 cases of p53 wildtype, 24 cases of p53 abnormal, and 18 cases of HPV-associated OED. Inter-pathologist agreement in OED grading according to 3-tiered (κ=0.32) and 2-tiered (κ=0.39) systems by H&E was poor, but fair-to-good (κ=0.59) in classification as p53 wildtype/p53 abnormal/HPV-associated by H&E and IHC. Classification of OED as p53 wildtype, p53 abnormal, or HPV-associated using p53/p16 IHC outperformed conventional grading in this cohort enriched for p53 abnormal OED, which required correct interpretation of p53 IHC, historically deemed challenging. Routine use of IHC also identifies a wider histologic spectrum of HPV-associated OED than is currently appreciated. More work is needed to determine the efficacy of this classification system in predicting patient outcomes and in guiding management decisions in real-world cohorts.

PURPOSE:Post prostatectomy PSA kinetics and General Grade Groups (GGG) are the strongest prognostic markers of biochemical recurrence (BCR) and prostate cancer (PCa)-specific mortality after radical prostatectomy. Despite having low-risk PCa, some patients will experience BCR, for some, clinically significant BCR. There is a need for an objective prognostic marker at the time of prostatectomy to improve risk stratification within this population. In this study, we investigated the prognostic potential of DNA ploidy.MATERIALS AND METHODS:Prostatectomy samples from 97 patients with GGG1 and GGG2 with a low-risk CAPRA-S score were included in this study. PCa tissue with the worst Gleason pattern underwent tissue disaggregation, cell isolation and staining with a DNA stoichiometric stain. Using image cytometry, DNA ploidy was measured and a Ploidy Score (PS) was generated.RESULTS:Among the 97 patients, 79 had no BCR, 18 experienced BCR, of which 14 had a PSA doubling time (PSA-DT) >1 year (low-risk group) and 4 had a PSA-DT of <1 year (high-risk group). Using Logistic regression analysis, only pathological T stage (pT) and PS independently predicted BCR with PS being the most significant (p = 0.001). The number of aneuploid cells was significantly higher in the high-risk group compared to the other groups (p = 1.7x10-11). PS combined with GGG diagnosis further stratified risk groups of biochemical recurrence free survival within CAPRA-S low-risk cohort.CONCLUSION:DNA ploidy is an independent prognostic marker of BCR in low-risk PCa after radical prostatectomy, which could early on identify potentially aggressive PCa recurrences and introduce a more personalized approach to salvage treatments.

Introduction Fanconi anemia (FA) is the most common cause of inherited bone marrow failure, and a cancer predisposition syndrome caused by impaired DNA interstrand crosslink repair, also known as the FA repair pathway. Hematopoietic stem cell transplantations (HSCT) have improved survival in patients with FA, however, those have also heightened the risk of malignancies. Head and neck squamous cell carcinoma (HNSCC), and specifically oral SCC (oSCC), are the most common solid malignancies and the leading cause of death in adult patients with FA. By 45 years of age, HNSCC has a computed cumulative incidence of 50% and 100% in FA patients who did not and did undergo a HSCT, respectively. Historically, SCC in FA has been mostly diagnosed at an advanced stage of disease conferring poor outcome. FA-associated HNSCC are known to be enriched with p53 and structural variants. We aim to develop molecular screening tools and biomarkers of oSCC development in FA patients, to facilitate measurement of potential prophylactic therapies and allow for earlier detection of premalignant oral lesions and cancer. To that end, we are mapping somatic genetic variants in normal-appearing oral mucosa as well as oral lesions in patients with FA, and we plan to evaluate circulating cancer biomarkers. Methods This is a proof-of-concept study conducted in healthy controls (HC) and FA patients. We use a non-invasive oral brush biopsy to collect oral keratinocytes from six different sites of normal-appearing mucosa: retromolar area, side of the tongue and floor of the mouth, bilaterally. DNA is isolated and the exonic regions of the TP53 gene are sequenced using CleanPlex TP53 Kit on an illumina Miseq platform, to identify single nucleotide variants (SNVs), insertion-deletions (indels) at an allele frequency above 1%. Genome-wide SNP genotyping with Infinium Global Diversity Array (v1.0) is used for copy number variant (CNV) identification. Areas of visible changes/abnormal appearing mucosa are also brushed and are tested for the aforementioned genetic changes, as well as for cytology and DNA ploidy. Circulating plasma and serum biomarkers will be evaluated at the end of the study. Results The study is ongoing. 20 HC and 17 FA patients have completed the study to date. Median age of the HC and FA cohorts were 36.3 years (range 23-61) and 31.8 years (range 15-50), respectively. Sex distribution was similar with males consisting of 44% and 53% in the HC and FA cohorts, respectively. The HC cohort included significantly more tobacco smokers and alcohol consumers than the FA cohort (33.3% vs. 0 for smokers, and 83.3% vs.17.6% for alcohol consumers, in the HC vs. FA cohorts, respectively, p<0.05). Using an optimized DNA isolation protocol, we were able to obtain sufficient amount of high molecular weight DNA, for each of the six sites brushed. Both freshly collected and stored samples were successfully used for molecular studies. Storing of HC samples in SurePath preservation solution, for an average duration of 55 days, did not impair DNA yield or sequencing data. No somatic molecular changes were evident in the oral mucosa of the healthy controls (n=18) in over 100 samples taken. At least one somatic molecular change was evident in most participants with FA. Five out of the 15 participants (33.3%) in the FA cohort tested positive for 10 different clinically significant somatic TP53 SNVs. At least one CNV was detected in samples sequenced from non-lesional oral mucosa in 10 out of 15 FA participants (66.7%), with numerous samples testing positive for multiple CNVs. The most frequently observed CNV was chromosome 9p isodisomy, observed in 6/15 participants (40%). Molecular analysis of 13 brushed lesions demonstrated 2 distinct clinically significant TP53 somatic variants in one lesion and multiple CNVs in 3 lesions, all from one individual with FA. Cytology and DNA ploidy analyses were negative in 2 lesions, the third lesion showed suspicious cytology, resulting in a surgical biopsy, which was negative. Analysis of additional samples and recruitment of FA participants is ongoing. Conclusions This study validates noninvasive screening applications for identification of somatic molecular changes in the normal-appearing mucosa in patients with FA. Further studies will assess if these changes can be used as surrogate biomarkers of a short-term outcome for the evaluation of preventative treatment effectiveness in FA-associated oSCC and early diagnosis.

OBJECTIVE:A disordered voice may significantly impair the ability of workers to perform optimally on the job, especially those classified as professional voice users (PVU). Voice therapy is a common treatment option for voice disorders, but there are few studies demonstrating its effect on work productivity. The objective of this study was to evaluate the change in work productivity in PVU after group voice therapy. STUDY DESIGN:Prospective cohort study SETTING: Academic Voice Center METHODS: PVUs whose primary treatment for their voice disorder was voice therapy were recruited. Participants completed a 7-week group voice therapy course from January 2018 to December 2020. Participants completed the validated Work Productivity and Activity Impairment questionnaire (WPAI) which measured presenteeism (on the job work productivity impairment) and absenteeism (time missed from work), general self-efficacy scale (GSES), and Voice handicap index -10 (VHI-10) before and after group voice therapy. Changes in scores before and after therapy were compared using a Wilcoxon Signed-Rank test. RESULTS:Twenty-seven PVU were recruited; 25 had complete data (100% female, mean age 45.4 years, 68% teachers). Presenteeism (SD) decreased from 72.0% (23.3) to 36.8% (24.8), which represented a significant improvement of 35.2% (27.8) [95% CI 21.7-38.7; P < 0.001]. Activity impairment decreased from 48.4% (32.0) to 25.6% (23.8), which represented a significant improvement of 22.8% (26.5) [95% CI 20.7-37.0; P < 0.00]. There was no change in absenteeism (P = 0.27). Patients had high mean GSES of 34.4 (3.7) and abnormal mean VHI-10 of 18.2 (7.2). Changes in VHI-10 and GSES were not significant. CONCLUSION:PVU had an improvement in work productivity that was largely represented by decreased presenteeism after completing group voice therapy.

BACKGROUND:Despite the oral cavity being readily accessible, oral cancer (OC) remains a significant burden. The objective of this study is to develop a DNA ploidy-based cytology test for early detection of high-risk oral lesions. METHODS:This retrospective study was conducted using 569 oral brushing samples collected from 95 normal and 474 clinically abnormal mucosa with biopsy diagnosis of reactive, low-grade or high-grade precancer or cancers. Brushing cells were processed to characterize DNA ploidy. A two-step DNA ploidy-based algorithm, the DNA ploidy oral cytology (DOC) test, was developed using a training set, and verified in test and validation sets to differentiate high-grade lesions (HGLs) from normal. The prognostic value of the test was evaluated by an independent outcome cohort, including progressed and non-progressing normal, reactive and low-grade lesions. Classification performance was assessed by accuracy, sensitivity, and specificity, while the prognostic value was evaluated by using the Cox proportional hazards analysis on 3-year progression-free survival (PFS). RESULTS:The developed DOC test exhibited high accuracy for detecting HGLs in the test and validation sets, with a sensitivity of 0.97 and 0.96, respectively. Its application to the Outcome cohort demonstrated significant prognostic value for 3-year PFS (log rank, p < 0.001). Multivariate analysis showed that high-grade pathology was the only variable explaining positive DOC test, not age, smoking, or lesional site. CONCLUSION:Clinical implementation of the DOC test could provide an effective screening method for detecting HGLs for biopsy and lesions at risk of progression.

Background Biomarkers that can identify which active surveillance (AS) prostate patients are likely to progress would have clinical utility. Methods Using a DNA specific stain (Feulgen-Thionin) and deep learning based segmentation to identify/delineate every nucleus in prostate needle biopsies with an accuracy comparable to human annotation even in areas of overlapping nuclei. Further quantifying the distribution of the DNA within these nuclei one can automatically define different tumour, immune and stromal cell categories. Within these cell categories one can identify subsets that construct classifiers that differentiate between AS patients that progress (35 cases, poor outcome) from those that do not (69 cases, good outcome). Approximately 5.4% cells across all cases were used to build these classifiers and they were tested on the remaining 94.6% of cells Results Across the 126 needle biopsies, ~1.5 million nuclei were segmented by the deep learning algorithm (>90% accuracy of correct segmentation upon visual examination). Classifier training involved 78,271 of these nuclei, leaving 1.4 million nuclei as test set. The frequency of immune like cells, density of tumour nuclei and subsets of tumour cells and stroma cells were found to be predictive of progression or non-progression. A combination of these cell frequencies defined risk groups that predict future behavior (See figure 1). Cases with a high frequency of immune like nuclei were predicted to have good outcome even if the features of the tumour cells predominately are predicted poor outcome. Examining the distribution of these immune cells within a prostate needle biopsies with spatial distinct tumour nuclei predicting poor outcomes and good outcomes it was interesting to observe that the immune cell density was much higher in the areas of tumour nuclei predicting poor outcomes. Suggesting that the immune system was also recognizing the aggressive poor outcome predicting cells and dealing with them as these patients did not progress. This also suggests that potentiating tumour immune cell interactions could potentially affect prostate patient outcomes. Conclusions Deep Learning segmentation paired with a DNA specific stain can identify/delineate cell nuclei with high accuracy and quantitative measures of the DNA distribution within these nuclei can be predictive of AS patient outcome. Immunes cell frequency identified in this fashion is predictive of patient outcomes. Ethics Approval Ethics approval was obtained from the UBC-BC Cancer Research Ethics Board. The number of the ethics approval is H1301398.

Abstract Previous evidence indicates that human papillomavirus (HPV) integration status may be associated with cervical cancer development and progression. However, host genetic variation within genes that may play important roles in the viral integration process is understudied. The aim of this study was to examine the association between HPV16 and HPV18 viral integration status and SNPs in nonhomologous-end-joining (NHEJ) DNA repair pathway genes on cervical dysplasia. Women enrolled in two large trials of optical technologies for cervical cancer detection and positive for HPV16 or HPV18 were selected for HPV integration analysis and genotyping. Associations between SNPs and cytology (normal, low-grade, or high-grade lesions) were evaluated. Among women with cervical dysplasia, polytomous logistic regression models were used to evaluate the effect of each SNP on viral integration status. Of the 710 women evaluated [149 high-grade squamous intraepithelial lesion (HSIL), 251; low-grade squamous intraepithelial lesion (LSIL, 310 normal)], 395 (55.6%) were positive for HPV16 and 192 (27%) were positive for HPV18. Tag-SNPs in 13 DNA repair genes, including RAD50, WRN, and XRCC4, were significantly associated with cervical dysplasia. HPV16 integration status was differential across cervical cytology, but overall, most participants had a mix of both episomal and integrated HPV16. Four tag-SNPs in the XRCC4 gene were found to be significantly associated with HPV16 integration status. Our findings indicate that host genetic variation in NHEJ DNA repair pathway genes, specifically XRCC4, are significantly associated with HPV integration, and that these genes may play an important role in determining cervical cancer development and progression. Prevention Relevance: HPV integration in premalignant lesions and is thought to be an important driver of carcinogenesis. However, it is unclear what factors promote integration. The use of targeted genotyping among women presenting with cervical dysplasia has the potential to be an effective tool in assessing the likelihood of progression to cancer.

The global burden of disease study revealed that lung cancer is a leading cause of death worldwide. The only patients with a prospect of cure are early non-small cell lung cancer (eNSCLC) amenable to surgery. However, many experience poor survival from recurrence and aggressive tumor microenvironment, which includes patient immune expression patterns undiagnosed with current clinical technology. Peri-operative immune checkpoint inhibitors (ICIs) provide promise to improve survival in eNSCLC. However, data on optimal immunohistohemistry (IHC) approach to select patients who will benefit from ICIs is needed.

Supplementary Tables 1, 2, 3: Table 1: Nuclear morphometric and texture features as calculated by the Getafics scanner (FB6 feature set description, Alexei Doudkine and Calum MacAulay, January 15, 1998). Table 2: Sample breakdown based on biopsy diagnosis and progression. Table 3: Comparison of high-risk and low-risk DNA brushing classification according to demographic, risk habit and lesion clinical features.

The growth and metastasis of solid tumours is known to be facilitated by the tumour microenvironment (TME), which is composed of a highly diverse collection of cell types that interact and communicate with one another extensively. Many of these interactions involve the immune cell population within the TME, referred to as the tumour immune microenvironment (TIME). These non-cell autonomous interactions exert substantial influence over cell behaviour and contribute to the reprogramming of immune and stromal cells into numerous pro-tumourigenic phenotypes. The study of some of these interactions, such as the PD-1/PD-L1 axis that induces CD8+ T cell exhaustion, has led to the development of breakthrough therapeutic advances. Yet many common analyses of the TME either do not retain the spatial data necessary to assess cell-cell interactions, or interrogate few (<10) markers, limiting the capacity for cell phenotyping. Recently developed digital pathology technologies, together with sophisticated bioimage analysis programs, now enable the high-resolution, highly-multiplexed analysis of diverse immune and stromal cell markers within the TME of clinical specimens. In this article, we review the tumour-promoting non-cell autonomous interactions in the TME and their impact on tumour behaviour. We additionally survey commonly used image analysis programs and highly-multiplexed spatial imaging technologies, and we discuss their relative advantages and limitations. The spatial organization of the TME varies enormously between patients, and so leveraging these technologies in future studies to further characterize how non-cell autonomous interactions impact tumour behaviour may inform the personalization of cancer treatment.​